| name | zh-hans-target-quality |
| description | Use when translating public-domain or licensed books into Simplified Chinese, regardless of source language. Provides target-language quality rules for AI-assisted translation, including Chinese readability, style research, chapter gates, benchmark testing, literary revision, and EPUB-ready final text checks. |
Simplified Chinese Target Quality / 简体中文目标语言质量
Use this skill before translating full chapters. Do not begin batch translation until a sample passes the quality gate.
Required Workflow
- Read
references/research_digest.md.
- Read
references/quality_standard.md.
- Read
references/workflow.md.
- Create
metadata/book_specific_translation_research.md with templates/book_specific_research_prompt.md.
- Create
metadata/style_profile.md with templates/style_profile_prompt.md.
- If the user provides lawful private benchmark excerpts, evaluate them with
templates/private_benchmark_compare_prompt.md.
- Run pretranslation trials with
templates/pretranslation_trial_prompt.md.
- Run the sample test described in
tests/benchmark_protocol.md and templates/sample_translation_test_prompt.md.
- Only after pretranslation and sample tests pass, translate chapters with
templates/chapter_translation_prompt.md.
- Revise failed chapters with
templates/revision_prompt.md.
- Audit image-bearing word choices with
templates/image_word_audit_prompt.md.
- Gate every chapter with
templates/chapter_quality_gate_prompt.md.
- Score every final chapter with
templates/evaluation_rubric.md.
Hard Rules
- Never optimize only for literal correspondence.
- Never keep English syntax when natural Chinese would say it differently.
- Never preserve a metaphor word-for-word when it becomes obscure or ugly in Chinese; preserve the effect instead.
- Never use copyrighted reference translations as source material unless the user provides lawful excerpts for private evaluation. Do not store long copyrighted excerpts in the project.
- If a sentence sounds like a translation, rewrite it unless the stiffness is intentional in the original.
- During the translation call itself, use a slim prompt and output only the translation. Do not ask the model to write QA, release notes, lint records, or workflow explanations in the same call.
- Natural Chinese is the first hard constraint in translation. Faithful but breathless long sentences, English clause skeletons, or compliance-looking prose that does not read like Chinese fail the gate.
- If a passage is historically sensitive, keep the historical fact and tone, but add translator notes instead of silently modernizing.
- Do not translate a chapter without a book-level style profile and a passing sample-test report.
- In body text, prefer accurate image-bearing words over flat explanatory words when the source is symbolic, scenic, physical, or memorable.
- Do not start batch chapter translation without a passing
qa/pretranslation/pretranslation_report.md.
Quality Gate
A chapter can move to chapters/final/ only if:
- No major meaning errors.
- No missing paragraph.
- No unresolved proper noun or technical term inconsistency.
- Readability score is at least 4/5.
- Literary/narrative effect score is at least 4/5.
- Chinese-only readability score is at least 4/5 before source calibration.
- A 20-sentence read-aloud sample has at most one clearly awkward sentence, and no key sentence is breathless.
- The reviewer cannot point to a sentence that feels mechanically translated.
- A gate report exists under
qa/gates/ and says PASS.
qa/samples/sample_test_report.md exists and says PASS.
qa/pretranslation/pretranslation_report.md exists and says PASS.
- An imagery audit exists under
qa/imagery/ for chapters with symbolic, scenic, physical, or memorable language.